Uveal Melanoma, Angiogenesis and Immunotherapy, Is There Any Hope?

Uveal melanoma is considered a rare disease but it is the most common intraocular malignancy in adults. Local treatments are effective, but the systemic recurrence rate is unacceptably high. Moreover, once metastasis have developed the prognosis is poor, with a 5-year survival rate of less than 5%, and systemic therapies, including immunotherapy, have rendered poor results. The tumour biology is complex, but angiogenesis is a highly important pathway in these tumours. Vasculogenic mimicry, the ability of melanomas to generate vascular channels independently of endothelial cells, could play an important role, but no effective therapy targeting this process has been developed so far. Angiogenesis modulates the tumour microenvironment of melanomas, and a close interplay is established between them. Therefore, combining immune strategies with drugs targeting angiogenesis offers a new therapeutic paradigm. In preclinical studies, these approaches effectively target these tumours, and a phase I clinical study has shown encouraging results in cutaneous melanomas. In this review, we will discuss the importance of angiogenesis in uveal melanoma, with a special focus on vasculogenic mimicry, and describe the interplay between angiogenesis and the tumour microenvironment. In addition, we will suggest future therapeutic approaches based on these observations and mention ways in which to potentially enhance current treatments.


Introduction
Uveal melanoma (UM) is a malignant tumour that arises in the melanocytes located in the uveal tract [1]. In 85-90% of cases the choroid is involved, while in the remaining 10-15% the tumour will arise in the iris and ciliary body [2][3][4]. It is considered a rare cancer, with an estimated incidence of 4.9-5.2 cases per million in the United States [5], which has remained stable in the past 20 years [6]. In Europe, the incidence seems to be highest in the Northern countries, with an incidence rate of more than 8 cases per million in Norway and Denmark [7]. However, in Southern countries, such as Spain  the bottom 50% (Low angiogenesis enrichment score). The cut-off used for generating high and low groups was the median enrichment score. The gene set used for scores was the same as for Figure 1. Data were extracted from the TCGA database through cBioPortal; 60 primary UM samples, 99 primary cutaneous melanoma samples.
The hypoxic microenvironment of these tumours leads to the release of hypoxia-inducible factor 1 (HIF-1), which in turn induces VEGF, Ang-2, matrix metalloprotease 14 (MMP14), and angiogenin A B Figure 1. Angiogenesis enrichment score comparing relapsed vs. non-relapsed tumours. Scores were generated from expression data by gene set variation analysis (Gene Set Variation Analysis (GSVA) function). The gene set "Biocarta VEGF Pathway" from MSigDatabase (which includes genes related to hypoxia, blood vessels formation, and pro-angiogenic factors) was used (http://software.broadinstitute. org/gsea/msigdb/cards/BIOCARTA_VEGF_PATHWAY). Groups were compared by non-parametric Wilcoxon rank test. Data were extracted from The Cancer Genome Atlas (TCGA) database through cBioPortal (TCGA-UVM, 80   Disease-free survival for uveal melanoma (UM) (A) and cutaneous melanoma (B) from patients included in the TCGA, clustered into the top 50% (high angiogenesis enrichment score) vs. the bottom 50% (Low angiogenesis enrichment score). The cut-off used for generating high and low groups was the median enrichment score. The gene set used for scores was the same as for Figure 1. Data were extracted from the TCGA database through cBioPortal; 60 primary UM samples, 99 primary cutaneous melanoma samples.
The hypoxic microenvironment of these tumours leads to the release of hypoxia-inducible factor 1 (HIF-1), which in turn induces VEGF, Ang-2, matrix metalloprotease 14 (MMP14), and angiogenin A B Figure 2. Disease-free survival for uveal melanoma (UM) (A) and cutaneous melanoma (B) from patients included in the TCGA, clustered into the top 50% (high angiogenesis enrichment score) vs. the bottom 50% (Low angiogenesis enrichment score). The cut-off used for generating high and low groups was the median enrichment score. The gene set used for scores was the same as for Figure 1. Data were extracted from the TCGA database through cBioPortal; 60 primary UM samples, 99 primary cutaneous melanoma samples. The molecular pathways involved in angiogenesis in melanoma are complex and beyond the scope of this review [85]. Notch1 seems to be increasingly important, especially in cutaneous melanoma. Indeed, in a review of 114 primary cutaneous melanoma carried out by Murtas et al. [86], the overexpression of Notch1 in both tumour and endothelial cells was associated to microvascular density. Notch1 seems to upregulate mitogen-activated protein kinase (MAPK) through CD133, which in turn transdifferentiates into endothelial-like phenotypes, therefore promoting growth and angiogenesis [87]. In UM, however, there is no established direct relationship between the Notch signalling pathway and angiogenesis. Nevertheless, hypoxia does seem to promote growth and invasion of uveal melanoma cell lines through the activation of Notch and MAPK [88,89].
Another important role of Notch1 seems to be its involvement in vascular mimicry [90], which we will discuss in the following section.

Vasculogenic Mimicry
Vasculogenic mimicry is referred to as the process by which aggressive melanoma cells generate vascular channels independently of endothelial cells [46]. This was first described by Maniotis et al. 1999 in cutaneous and UM tissue sections [91]. The authors observed interconnected loops of extracellular matrix containing some red blood cells, with no evidence of endothelial cells. These patterns were more common in highly invasive melanomas, whereas normal melanocytes or poorly invasive cells were unable to generate such channels [91]. In 2008, Frenkel et al. used laser scanning confocal angiography with indocyanine green to demonstrate blood circulation through leakage [92]. Since this initial description, vasculogenic mimicry has been found in many other tumours and seems to be correlated to poor tumour differentiation, lymph node involvement, distant metastasis, and TNM stage [93], and therefore entails a decreased survival [93,94].
The molecular processes and mechanisms involved in vasculogenic mimicry are elusive, and much remains to be understood [95,96]. The process is triggered by a reversion of the melanocytic tumour cells to a pluripotent embryonic-like genotype [91,97,98]. The acquisition of these stem-like properties seems to be mediated by the induction of the epithelial-mesenchymal transition (EMT), as initially observed in mammary epithelial cells [99]. Indeed, coexpression of epithelial and mesenchymal markers has been observed in cutaneous melanoma cells engaged in vasculogenic mimicry [100,101]. Additionally, a significant overexpression of EMT transcription factors involved in both the acquisition of stem cell-like properties and vasculogenic mimicry have been observed in different tumours, such as Nodal in murine melanoma and human cutaneous melanoma [102][103][104][105], Twist in hepatocellular carcinoma [106][107][108], Bcl-2 in human melanoma and hepatocellular carcinoma [108,109], Zinc-finger E-box binding homeobox (ZEB) in hepatocellular, pancreatic, and colorectal carcinoma, [110][111][112], or Snail in human breast and oral squamous cell carcinomas [113,114], amongst others. Despite this relationship not being directly established in UM, the expression of EMT-associated factors does promote invasion and growth [115], which we believe could be partly explained by their role in vasculogenic mimicry, as has been demonstrated in other tumours. Only a small subset of cells, globally termed melanoma cancer stem cells (MCSCs), that are preferentially localized in the perivascular niches of cutaneous melanomas and express stem cell markers, seem to be involved in vasculogenic mimicry [116].
MCSCs involved in the formation of tubules in cutaneous melanoma cell lines highly express vascular endothelial-cadherin (VE-cadherin) [117], a central actor in vasculogenic mimicry. Indeed, downregulation of this molecule in both cutaneous and UM lines completely abrogates vasculogenic mimicry [117,118]. VE-cadherin is induced during EMT transition [119] and upregulates transforming growth factor β (TGF-β) in breast cancer cells. Additionally, VE-cadherin is colocalized with EphA2 at cell-cell adhesions and regulates EphA2 at the cell membrane in both uveal and cutaneous melanoma lines, by mediating its ability to become phosphorylated through interactions with its membrane bound ligand, ephrin-A1 [120]. The exact molecular pathways involved in the regulation of this complex interplay remain unknown, although these findings, along with others [121] evidence the importance of intracellular phosphorylation in vasculogenic mimicry.
Microarray gene chip analysis has revealed increased expressions of laminin 5, membrane type 1-matrix metalloproteinases (MT1-MMP), and MMP-1, -2, -9, and -14 in aggressive metastatic melanoma cells compared to poorly aggressive ones, and the inhibition of the interaction between laminin-5 and MMP-2 and MT1-MMP with specific antibodies inhibits the formation of the tubular network, suggesting a strong implication of these components in vasculogenic mimicry [122]. The overexpression of MT1-MMP and MMP-2 is regulated by phosphoinositide 3-kinase (PI3K), and specific inhibitors of PI3K are able to abrogate vasculogenic mimicry in both uveal and cutaneous melanoma cells by decreasing the levels of MT1-MMP and MMP-2 [123].
Furthermore, the hypoxic microenvironment of melanomas seems to be an additional trigger of vasculogenic mimicry [124]. Mouse melanoma B16 cells implanted in the ischemic limbs of mice have more vasculogenic mimicry channels than controls, with a higher expression of HIF-1, MMP-2, MMP-9, and VEGF [125,126]. Indeed, vasculogenic mimicry seems to be partly mediated by vascular endothelial growth factor receptor 1 (VEGFR-1), as was shown by Frank et al. in ABCB5+ melanoma xenografts [127]. However, a second, VEGF-independent mechanism is also able to trigger vasculogenic mimicry, mediated by the platelet EC adhesion molecule (PECAM-1), whose expression is repressed by the neural crest specifier AP-2α [128].
The relative contribution of the aforementioned mechanisms of tumour blood supply to disease progression are unknown. However, in a study performed in mouse melanoma xenographs, mosaic vessels, vasculogenic mimicry, and endothelium-dependent vessels were observed in all stages of tumour development. However, vasculogenic mimicry seemed to be the predominant pattern in early stages of disease, while this was replaced by endothelium-dependent vessels in later stages of disease development [129]. In intraocular melanoma models, the three types of microcirculation were also observed, but endothelium-dependent vessels were more common in larger tumours while vascular mimicry seemed to be predominant in smaller lesions [130]. Therefore, vasculogenic mimicry seems to play a predominant role in early stages of disease development, both in cutaneous and UM. These observations seem to be contradicted by the initial findings of Maniotis et al. and Chang et al. [91,131], where larger tumours seemed to be richer in matrix-embedded channels. Whether vasculogenic mimicry is a time-dependent event in disease progression or simply identifies inherently more aggressive tumours is unclear.
Folberg et al. demonstrated that UM that presented vasculogenic mimicry patterns had an upregulation of genes related to differentiation and suppression of proliferation, and a downregulation of genes related to promotion of invasive and metastatic behaviour [132]. These findings are counterintuitive given the worse prognosis of melanoma patients presenting with his pattern. However, the authors hypothesize that these findings could explain the chemoresistance observed in these tumours and the late metastatic recurrences of some of these patients.

Antiangiogenic Drugs in Uveal Melanoma
Based on the preclinical evidence previously discussed, targeting angiogenesis in UM seems to be an attractive and potentially effective strategy [133]. However, results of the clinical trials that have investigated this matter have been disappointing so far (see Table 1), although they do seem to be more active than in cutaneous melanoma. Most antiangiogenic drugs render no response [134][135][136][137][138][139] and in the largest trial performed to date, the best observed response rate with sorafenib was 1.7% [140]. However, cabozantinib showed higher PFS in uveal melanoma patients compared to cutaneous melanoma [139]. This observation is based on indirect comparisons and must therefore be taken with caution. Some antiangiogenic drugs do seem to be able to produce disease stabilization in more than 50% of patients [134,[139][140][141][142], albeit for a short period of time in most cases. Moreover, one must bear in mind that the number of patients included in these trials is low, with less than 20 patients in most cases, which hinders the generalizability of the data (see Table 1). Some have questioned whether the stabilizations observed were in fact due to the antiangiogenic effects of these drugs or were simply reflecting the natural history of this disease. In order to try to answer this question, Scheulen et al. carried out the STEAM study, a randomized phase II trial, in which all patients were initially treated with sorafenib for 56 days in a run-in period. Patients with stable disease were then randomized to continue on sorafenib or placebo. The trial demonstrated a significant increase in progression-free survival (5.5 vs. 1.9 months, Hazard Ratio (HR) 0.527, p = 0.0079) in patients that continued on sorafenib. Cross-over to sorafenib was allowed following progression to placebo. That might explain that there were no differences observed in overall survival (OS).
Therefore, new approaches are needed in order to optimize outcomes with antiangiogenic therapies in UM.
One of the main concerns in the current approaches is that antiangiogenic strategies target molecules that are mainly involved in endothelium-mediated angiogenesis, leaving vasculogenic mimicry aside. However, as we have discussed, vasculogenic mimicry seems to be present in more aggressive melanomas, and although its biological implication is unclear, its inhibition could have a potential therapeutic role. Indeed, van der Schaft et al. demonstrated that using three angiogenic inhibitors (anginex, TNP-470 and endostatin) in human melanoma as compared to human endothelial cell lines inhibited angiogenesis in endothelial lines but did not inhibit vasculogenic mimicry in the melanoma lines. A further analysis revealed a differential expression of two endostatin receptors that could explain these observations [145]. Targeting vasculogenic mimicry, therefore, should be pursued and has been attempted with different approaches in preclinical melanoma models: • Genistein, an isoflavone present in soybeans, is able to inhibit vasculogenic mimicry in uveal melanoma C918 cell lines by reducing the expression of VE-cadherin [146].
• Pevonedistat, a selective and potent inhibitor of NEDD8-activating enzyme E1 subunit 1 (NAE1), an enzyme involved in neddylation, is able to repress the cancer stemness properties of UM cell lines, and could therefore potentially interfere with vasculogenic mimicry [147]. • Fasudil, a Rho kinase inhibitor, is able to reduce tumour growth in melanoma cell lines and melanoma mice models by inhibiting vasculogenic mimicry [148]. • Nicotinamide, the amide form of vitamin B3 (niacin), effectively targets vasculogenic mimicry by downregulating VE-cadherin in cutaneous melanomas cell lines. However, melanoma cells seemed to acquire an increased invasion capacity [149]. • A chemically modified tetracycline is able to inhibit MMP-2 and -9 in addition to laminin 5 in both cutaneous and uveal melanoma cell lines. Moreover, the expression of vasculogenic mimicry-associated genes is also inhibited [150]. • Cilenglitide, a potent inhibitor of αv integrins activation, reduces extracellular matrix invasion, vasculogenic mimicry, and secretion of MM9 by selectively targeting αvβ5 integrin in human cutaneous melanoma cell lines [151].

•
Thalidomide, a drug with antiangiogenic and immunomodulatory properties, is able to decrease the number of vasculogenic mimicry tubules and reduce the protein expression of MMP-2, MMP-9, VEGF, proliferating cell nuclear antigen (PCNA), and nuclear factor-κβ (NF-κβ) in murine cutaneous melanoma cell lines [152]. • PARP inhibitors suppress the metastatic potential of some human and murine melanoma cells, due in part to the inhibition of vasculogenic mimicry mediated by the downregulation of VE-cadherin and the inhibition of the EMT pathway [153].

•
Novel molecules, such as CVM-1118, seem to inhibit vasculogenic mimicry by targeting essential pathways involved in the process in human melanoma cells [154].
Targeting pericytes is another plausible therapeutic opportunity [155] and could help overcome tumour resistance to other drugs [156]. The proteoglycan NG2 stimulates the proliferation, motility, and migration of pericytes, and is crucial in the early stages of neovascularization [157]. Inhibition of pericytes via NG2 in UM xenografts decreases neovascularization and tumour volume, thereby rendering it a potential target [158,159].
Another possible approach would be a combination of antiangiogenic drugs with other therapies. For instance, combining bevacizumab with radiotherapy in UM mice models was shown to significantly decrease tumour growth compared to either bevacizumab or radiotherapy alone [160]. Combining antiangiogenic drugs with immunotherapy could have a potentially synergistic effect, as we will discuss in the following section.

The Potential Benefit of Combining Antiangiogenic and Immune Strategies
UM has traditionally been considered an immune privileged tumour, in a manner that closely parallels the microenvironment of the eye [161,162]. However, immune infiltration is found in the primary tumour [163,164], and its presence is associated to decreased survival [164], contrary to what is observed in other tumours [165]. Interestingly, metastatic UM seems to have a different lymphocytic composition compared to metastatic cutaneous melanoma, with predominant CD4+ lymphocytes instead of CD8+ lymphocytes (more commonly observed in cutaneous melanoma) [166]. Moreover, in contrast to most other tumours, the cytotoxic phenotype does not seem to confer better prognosis in UM [167], possibly reflecting the dominant immunosuppressive microenvironment [168,169]. UM also substantially differs from cutaneous melanoma in other aspects. For instance, the tumour mutation burden (TMB) (a surrogate for tumour antigenicity) of cutaneous melanoma is one of the highest among all tumours studied by the TCGA, whereas UM shows one of the lowest (Figure 3). However, when considering CD8A and PDL1 expression (a surrogate for tumour immunogenicity), almost 50% of cutaneous melanomas are considered CD8A high /PDL1 high , whereas the same percentage of UM are considered CD8A low /PDL1 low , reflecting a more immunosuppressed tumour microenvironment  Figure 4A,B). Interestingly the behaviour of UM and cutaneous melanoma in the highest and lowest CD8A/PDL1 quart differs substantially. Interestingly, CD8A low /PDL1 low cutaneous melanoma show a dismal prognosis, whereas CD8A high /PDL1 high UM convey a worse prognosis ( Figure 4C). surrogate for tumour immunogenicity), almost 50% of cutaneous melanomas are considered CD8A high /PDL1 high , whereas the same percentage of UM are considered CD8A low /PDL1 low , reflecting a more immunosuppressed tumour microenvironment ( Figure 4A,B). Interestingly the behaviour of UM and cutaneous melanoma in the highest and lowest CD8A/PDL1 quart differs substantially. Interestingly, CD8A low /PDL1 low cutaneous melanoma show a dismal prognosis, whereas CD8A high /PDL1 high UM convey a worse prognosis ( Figure 4C).   surrogate for tumour immunogenicity), almost 50% of cutaneous melanomas are considered CD8A high /PDL1 high , whereas the same percentage of UM are considered CD8A low /PDL1 low , reflecting a more immunosuppressed tumour microenvironment ( Figure 4A,B). Interestingly the behaviour of UM and cutaneous melanoma in the highest and lowest CD8A/PDL1 quart differs substantially. Interestingly, CD8A low /PDL1 low cutaneous melanoma show a dismal prognosis, whereas CD8A high /PDL1 high UM convey a worse prognosis ( Figure 4C).   A substantial number of UMs show no lymphocytic infiltration on histologic sections [164]. The immune infiltrate in UM seems to be related to genetic alterations, with the lack of BAP1 mutations showing a richer T-cell infiltration [170]. Another mechanistic explanation for this is the lack of adhesion of lymphocytes to the newly formed vessels [171]. Indeed, leukocytes require a number of molecules in order to roll, adhere, and finally transmigrate into the tumour microenvironment, including selectins, PECAM-1, intracellular adhesion molecular-1, -2 (ICAM-1, ICAM-2, respectively), and vascular cell adhesion molecular -1 (VCAM-1), amongst others [172]. Intratumoural cutaneous melanoma vessels have a decreased expression of P-selectin, VCAM-1, E-selectin, and ICAM-1, although the normal adjacent tissues have normal expression of these molecules [173][174][175][176]. The downregulation of these adhesion molecules could be mediated by the overexpression of VEGF [176]. Additionally, the tumour endothelium expresses molecules that can block T-cell infiltration in different tumour types (including melanoma), such as FasL [177], endothelin B receptor (ETBR), and endothelin-1 [178].
In addition to the expression of certain adhesion molecules on the endothelium, a chemoattractant is necessary to correctly recruit lymphocytes, especially chemokines that signal through the C-C chemokine receptor 5 (CCR5) and C-X-C Receptor 3 (CXCR3) axes [171]. Indeed, the co-expression of both molecules leads to a high CD8+ T-cell infiltration in cutaneous melanoma [179,180], and upregulation of both CCR5/CXCR3 is associated to greater response to different immunotherapies, including checkpoint inhibitors [181] and adoptive cell therapy [182]. Interestingly, the downregulation of these molecules or their corresponding ligands is correlated to disease progression [173,183].
Little is known about the interaction of vessels formed during vasculogenic mimicry and leukocytes. Nevertheless, studies have shown that the expression of some adhesion molecules on these vessels, such as PECAM-1 [128], possibly allowing circulating cells to interact with these tumour vessels. This raises the possibility of the tumour being able to regulate its own microenvironment by recruiting specific immune cells [171]. This area of research is currently under active investigation.
Highly angiogenic melanomas are more resistant to checkpoint inhibitors [38], probably due to the close relationship between aberrant cancer angiogenesis and immunosuppression [184]. Indeed, the tumour microenvironment, often characterized by hypoxia and high interstitial fluid pressure [185], could not only enhance an immunosuppressive microenvironment but also reduce the effectiveness of immunotherapy [37]. Moreover, VEGF (along with other angiogenic factors) plays a crucial role in modulating the immune system and fostering an immunosuppressive microenvironment: it directly suppresses dendritic cell maturation breast and colon carcinomas [39,186], inhibits T-cells by enhancing PD-1 and other inhibitory checkpoints in colon carcinomas [40,187], disrupts the normal differentiation of haematopoietic precursor cells [188], and recruits immunosuppressive cells such as T-cells [189] and myeloid derived suppressor cells [41,190].
Therefore, selectively targeting VEGF could not only inhibits angiogenesis but also change the tumour microenvironment, making it more "immunoresponsive" [185]. However, a more judicious use of antiangiogenic therapies would not only target angiogenesis but could give rise to tumour vessels with structural and functional phenotypes that are closer to non-malignant tissues, a process commonly referred to as vascular normalization [191]. This could subsequently result in an increased accumulation of cytotoxic T-cells [192], thereby improving the efficacy of checkpoint inhibitors.
The strategy of combining immune-checkpoint inhibitors with antiangiogenic drugs has been studied in other tumours [193]. For instance, three recent phase III studies combining anti-PD1/PDL1 with antiangiogenic therapies showed promising results in advanced renal cell carcinoma [194,195]. Pembrolizumab and axitinib increased overall survival compared to standard first-line sunitinib [195], and the combination of avelumab-axitinib increased progression-free survival [194]. Atezolizumab-bevacizumab also showed an increase in progression-free survival, although the survival data were still immature. In non-small cell lung cancer, the combination of chemotherapy with bevacizumab and atezolizumab increased progression-free survival and overall survival compared to chemotherapy and bevacizumab alone [196,197].
Based on these preclinical observations, Hodi et al. 2014 carried out a phase I clinical trial in patients with advanced cutaneous melanoma that received a combination of ipilimumab and bevacizumab. They observed a disease control rate of 64.7% and an OS of 25.1 months. More interestingly, however, on-treatment tumour biopsies revealed activated vessel endothelium with increased expression of E-selectin and increased infiltration of CD8+ T-cells [198]. Although angiogenesis seems to confer worse prognosis to UM when compared to cutaneous melanoma, to the best of our knowledge no clinical trial has been undertaken to study the benefits of combining antiangiogenic therapies with immunotherapy. The Grupo Español de Melanoma (GEM), a Spanish collaborative group, has recently designed a phase II, single-arm study in patients with metastatic UM who will be treated with the combination of durvalumab, an antiPD-L1 inhibitor, and cediranib, a multikinase inhibitor of VEGFR, PDGFRβ, KIT, and FLT-1 and -2. The primary endpoint is to evaluate the efficacy and response rate of the combination of cediranib and durvalumab in patients with metastatic UM with biopsiable disease at baseline, in first line or after failure to first-line systemic or liver-directed therapies. Five centres will be involved, and 18 patients are expected to be included, although the total number of patients may increase to 27 if the ORR > 20%.

Conclusions
In summary, angiogenesis plays an essential role in the development and progression of UM, and the exact implication of vasculogenic mimicry is still unclear, but could potentially play an important role. The efficacy of antiangiogenic drugs is still insufficient. Given the close relationship between angiogenesis and the immune microenvironment, combining antiangiogenic therapies and checkpoint inhibitors offers a potentially groundbreaking strategy. Nevertheless, we believe a deeper knowledge of vasculogenic mimicry is urgently needed. As we have seen, no effective strategy has been developed thus far and current antiangiogenic therapies not only do not target this pathway but could even overactivate it. This, in turn, could render antiangiogenic combination therapies ineffective. Moreover, targeting vasculogenic mimicry could not only contribute to the normalization of blood supply but also modulate the tumour microenvironment, therefore rendering immune therapies more effective.

Conflicts of Interest:
The authors declare no conflict of interest.